DexEXO: A Wearability-First Dexterous Exoskeleton for Operator-Agnostic Demonstration and Learning

Alvin Zhu; Mingzhang Zhu; Beom Jun Kim; Jose Victor S. H. Ramos; Yike Shi; Yufeng Wu; Raayan Dhar; Fuyi Yang

cs.ROMar 18, 2026

DexEXO: A Wearability-First Dexterous Exoskeleton for Operator-Agnostic Demonstration and Learning

Authors: Alvin Zhu, Mingzhang Zhu, Beom Jun Kim, Jose Victor S. H. Ramos, Yike Shi, Yufeng Wu, Raayan Dhar, Fuyi Yang, +5 more

Organizations: Department of Computer Science, UCLA, Los Angeles, CA, USA. · Department of Electrical and Computer Engineering, UCLA, Los Angeles, CA, USA. · Department of Mechanical and Aerospace Engineering, UCLA, Los Angeles, CA, USA.

Abstract

Scaling dexterous robot learning is constrained by the difficulty of collecting high-quality demonstrations across diverse operators. Existing wearable interfaces often trade comfort and cross-user adaptability for kinematic fidelity, while embodiment mismatch between demonstration and deployment requires visual post-processing before policy training. We present DexEXO, a wearability-first hand exoskeleton that aligns visual appearance, contact geometry, and kinematics at the hardware level. DexEXO features a pose-tolerant thumb mechanism and a slider-based finger interface analytically modeled to support hand lengths from 140~~mm to 217~~mm, reducing operator-specific fitting and enabling scalable cross-operator data collection. A passive hand visually matches the deployed robot, allowing direct policy training from raw wrist-mounted RGB observations. User studies demonstrate improved comfort and usability compared to prior wearable systems. Using visually aligned observations alone, we train diffusion policies that achieve competitive performance while substantially simplifying the end-to-end pipeline. These results show that prioritizing wearability and hardware-level embodiment alignment reduces both human and algorithmic bottlenecks without sacrificing task performance. Project Page: https://dexexo-research.github.io/

Figures & tables

Fig. 1 : System overview of DexEXO. (a) Full device worn on a user’s hand. (b) Demonstrations of piano playing, full-hand grasping, and scissors cutting. (c) Policy deployment on the robot.

Fig. 2 : Mechanical overview of DexEXO. DexEXO integrates a linkage-driven wearable exoskeleton, a passive data-capture hand, and an onboard sensing/power module. Insets highlight key subsystems: (a) passive finger slider for cross-user fit, (b) pose-tolerant thumb coupling interface, (c) parallel four-bar finger linkage for motion transmission, and (d) passive hand thumb that reproduces the intended thumb DOF

Fig. 4 : An overview of the full demonstration data modalities, policy training, and inference with visual-aligned observations.

	Scissors Cutting		Page Flipping		Cup Stacking		Piano Playing
Method	Success Rate	Time (s) †	Success Rate	Time (s) †	Success Rate	Time (s) †	Success Rate	Time (s) †
DexEXO	$\mathbf{0.79\pm 0.10}$	$\mathbf{11.7\pm 1.4}$	$\mathbf{0.88\pm 0.03}$	$5.4\pm 0.6$	$\mathbf{0.82\pm 0.07}$	$12.0\pm 1.1$	$\mathbf{0.96\pm 0.02}$	$\mathbf{21.6\pm 1.8}$
DexUMI	$0.00\pm 0.00$	—	$0.86\pm 0.04$	$\mathbf{4.7\pm 0.7}$	$0.80\pm 0.07$	$\mathbf{8.9\pm 1.0}$	$0.62\pm 0.13$	$25.9\pm 2.5$
Teleoperation	$0.00\pm 0.00$	—	$0.51\pm 0.06$	$18.0\pm 2.1$	$0.33\pm 0.09$	$68.6\pm 13.1$	$0.60\pm 0.09$	$97.4\pm 7.8$

TABLE I : Summary of quantitative results in user study (mean

\pm

SEM). Bold indicates best performance per metric.

Fig. 6 : Wiggle-space evaluation of the thumb interface. (a) and (b) Two views of the measured wiggle space. The fitted ellipsoid represents the covariance envelope of the sampled trajectory. The red point indicates the nominal fingertip position within the ellipsoid. (c) Experimental setup with reflective markers attached for motion-capture measurement.

Fig. 7 : Policy rollouts for the block pick and place, egg carton, and bottle tasks.

Method	Tasks
Finger Condition	Block	Carton	Bottle
No	0.90	0.90	0.85
Yes	0.85	0.95	0.80

TABLE II : Policy evaluation comparisons across methods and tasks.

Explore similar work

From Wearable Interfaces to Dexterous Policies: Contact Shifts and Tactile Representations
Oct 6, 2026Ruitong Tian, Fang Xu, Noah B. Wilson +2Robotic Data Collection Robot Teleoperation
Department of Civil and Coastal Engineering, University of Florida, Gainesville, FL, USA. · Department of Mechanical and Aerospace Engineering, University of Florida, Gainesville, FL 32611, USA.
SEED-UMI: Sharing the Exoskeleton between human and robot for onE-to-one Dexterous demonstration
Sep 10, 2026Tengbo Yu, Jiahao Wu, Daohan Li +4Contact-Rich Robotic Manipulation Robotic Data Collection
State Key Laboratory of General Artificial Intelligence, School of Intelligence Science and Technology, Peking University · Delta Intelligence
RealDexUMI: A Wearable Universal Manipulation Interface for Dexterous Robot Learning
Jun 4, 2026Chaoyi Xu, Yixuan Jiang, Jiahui Huan +7Dexterous Manipulation In-Hand Manipulation
Peking University · BeingBeyond · Beihang University +2

Oct 6, 2026J/K move · Enter open · S save

From Wearable Interfaces to Dexterous Policies: Contact Shifts and Tactile Representations

Ruitong Tian, Fang Xu, Noah B. Wilson, Xianyao Li+1

Department of Civil and Coastal Engineering, University of Florida, Gainesville, FL, USA. · Department of Mechanical and Aerospace Engineering, University of Florida, Gainesville, FL 32611, USA.

Robotic Data Collection Robot Teleoperation Robot Imitation Learning Tactile Sensing for Robotic Manipulation

Open paper AskPDF arXiv

Unlike conventional teleoperation, wearable interfaces allow operators to collect dexterous demonstrations through their own hand motions while directly interacting with task objects. This direct interaction reduces dependence on the target robot during collection, but it also makes the collection hardware part of the physical process that generates each demonstration. Interface geometry can influence both how a task is performed and what tactile observations are recorded for learning. We study two versions of a DexUMI-family exoskeleton that share the same robot command definition, mapping procedure, and tactile module type but differ in hand-side geometry. The revised interface reduces reported physical demand, improves selected device ratings, enables tactile access in a precision grasp that is mechanically blocked by the baseline, and produces task-dependent changes in recorded contact. Lid twisting primarily exhibits a change in contact location, whereas egg carton opening primarily exhibits a change in contact occurrence. We then train matched policies using either a binary aggregate input or a spatial-plus-force input. On lid twisting and egg carton opening, policies trained on revised-interface demonstrations achieve higher success than those trained on baseline demonstrations, with a significant pooled interface effect. Across these and two additional tasks, USB insertion and soldering tool pick and place, the spatial-plus-force input likewise outperforms binary aggregation. These results show that wearable collection hardware is part of the data-generation process for robot learning. Such interfaces should therefore be evaluated not only through operator experience, but also through the tactile interactions they make recordable and the policy performance their demonstrations support.

Abstract

Explore similar work

From Wearable Interfaces to Dexterous Policies: Contact Shifts and Tactile Representations

SEED-UMI: Sharing the Exoskeleton between human and robot for onE-to-one Dexterous demonstration

RealDexUMI: A Wearable Universal Manipulation Interface for Dexterous Robot Learning